M. Horsch, D. Poole, Flexible policy construction by information refinement, in: Proc. 12th Conference on Uncertainty in Artificial Intelligence, Portland, OR, Morgan Kaufmann, San Francisco, CA, 1996, pp. 315-- 324.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... In the realm of action under uncertainty, Heckerman, et al. investigated methods for the optimal compilation of actions given a finite quantity of memory [35] Horsch and Poole explored the offline construction and incremental refinement of trees representing decision policies for real time action [39]. In related work, Horvitz explored issues and opportunities with the ideal precomputation and caching of platform results partial solutions to potential future real time domain level and metareasoning problems to enhance the overall performance of a system [47,49] Zilberstein and Russell ....

M. Horsch, D. Poole, Flexible policy construction by information refinement, in: Proc. 12th Conference on Uncertainty in Artificial Intelligence, Portland, OR, Morgan Kaufmann, San Francisco, CA, 1996, pp. 315-- 324.

....address: Intelligent Systems Lab, School of Computing Science, Simon Fraser University, Burnaby, B.C. Canada V5A 1S6. Email: mhorsch cs.sfu.ca In this paper, we study a particular anytime algorithm for the problem of constructing policies for decision problems represented as influence diagrams [Horsch Poole, 1998; Horsch, 1998] This algorithm has a number of general features: the optimal solution is not known before it is computed; the current best solution is incrementally improved, although it is not known in advance how much improvement will be gained by a single computational step; the value of the ....

....Systems Lab, School of Computing Science, Simon Fraser University, Burnaby, B.C. Canada V5A 1S6. Email: mhorsch cs.sfu.ca In this paper, we study a particular anytime algorithm for the problem of constructing policies for decision problems represented as influence diagrams [Horsch Poole, 1998; Horsch, 1998] This algorithm has a number of general features: the optimal solution is not known before it is computed; the current best solution is incrementally improved, although it is not known in advance how much improvement will be gained by a single computational step; the value of the current best ....

[Article contains additional citation context not shown here]

Horsch, M. C. 1998. Flexible Policy Construction by Information Refinement. Ph.D. Dissertation, Department of Computer Science, University of British Columbia.

....the extension which maximizes the increase in expected utility is called the maximal extension strategy. We have also implemented a greedy strategy which chooses the first extension it can find which increases the value of the policy. These strategies and heuristics are discussed in more detail in [Horsch, 1998] . 2 RANDOM ACCESS REFINEMENT: AN ANYTIME ALGORITHM In this section, we present an anytime algorithm for computing policies for multi stage influence diagrams. A policy is represented by a collection of decision trees, one for each decision node in the influence diagram. As in Section 1.3, ....

Horsch, M. C. 1998. Flexible Policy Construction by Information Refinement. Ph.D. Dissertation, Department of Computer Science, University of British Columbia. Forthcoming.

....algorithm is sufficiently general to make use of existing tools for probabilistic reasoning, and has already provided reasonably valuable (but non optimal) policies for influence diagrams with about states. The algorithm is an extension of the iterative refinement technique presented in [Horsch Poole, 1996] , applied to multi stage influence diagrams. The refinement is applied to the decision nodes in random access ordering (as opposed to the sequential ordering of dynamic programming) This paper is organized as follows. First we briefly discuss influence diagrams and the decision tree ....

....by decision trees very succinctly. 1.3 THE SINGLE STAGE ALGORITHM The single stage information refinement algorithm constructs a decision tree for a influence diagram with a single decision node. The following description is a brief synopsis. The algorithm has been described in more detail in [Horsch Poole, 1996] , and is similar to algorithms described in [Heckerman, Breese, Horvitz, 1989; Lehner Sadigh, 1993] For a given leaf N in a decision tree, its context G L is extensible if it does not contain all the information variables. We refer to the information variables which are not in the ....

[Article contains additional citation context not shown here]

Horsch, M. C., and Poole, D. 1996. Flexible policy construction by information refinement. In Proceedings of the Twelfth Conference on Uncertainty in Artificial Intelligence, 315--324.

....address: Intelligent Systems Lab, School of Computing Science, Simon Fraser University, Burnaby, B.C. Canada V5A 1S6. Email: mhorsch cs.sfu.ca In this paper, we study a particular anytime algorithm for the problem of constructing policies for decision problems represented as influence diagrams [Horsch Poole, 1998; Horsch, 1998] This algorithm has a number of general features: the optimal solution is not known before it is computed; the current best solution is incrementally improved, although it is not known in advance how much improvement will be gained by a single computational step; the value of the ....

....Systems Lab, School of Computing Science, Simon Fraser University, Burnaby, B.C. Canada V5A 1S6. Email: mhorsch cs.sfu.ca In this paper, we study a particular anytime algorithm for the problem of constructing policies for decision problems represented as influence diagrams [Horsch Poole, 1998; Horsch, 1998] This algorithm has a number of general features: the optimal solution is not known before it is computed; the current best solution is incrementally improved, although it is not known in advance how much improvement will be gained by a single computational step; the value of the current best ....

[Article contains additional citation context not shown here]

Horsch, M. C. 1998. Flexible Policy Construction by InformationRefinement. Ph.D. Dissertation, Department of Computer Science, University of British Columbia.

No context found.

Michael C. Horsch and David Poole. Flexible policy construction by information refinement. In Proceedings of the Twelfth Conference on Uncertainty in Artificial Intelligence, pages 315--324, 1996.

....the extension which maximizes the increase in expected utility is called the maximal extension strategy. We have also implemented a greedy strategy which chooses the first extension it can find which increases the value of the policy. These strategies and heuristics are discussed in more detail in [Horsch, 1998] . 2 RANDOM ACCESS REFINEMENT: AN ANYTIME ALGORITHM In this section, we present an anytime algorithm for computing policies for multi stage influence diagrams. A policy is represented by a collection of decision trees, one for each decision node in the influence diagram. As in Section 1.3, ....

Horsch, M. C. 1998. Flexible Policy Construction by InformationRefinement. Ph.D. Dissertation, Department of Computer Science, University of British Columbia. Forthcoming.

....algorithm is sufficiently general to make use of existing tools for probabilistic reasoning, and has already provided reasonably valuable (but non optimal) policies for influence diagrams with about 2 61 states. The algorithm is an extension of the iterative refinement technique presented in [Horsch Poole, 1996] , applied to multi stage influence diagrams. The refinement is applied to the decision nodes in random access ordering (as opposed to the sequential ordering of dynamic programming) This paper is organized as follows. First we briefly discuss influence diagrams and the decision tree ....

....by decision trees very succinctly. 1.3 THE SINGLE STAGE ALGORITHM The single stage information refinement algorithm constructs a decision tree for a influence diagram with a single decision node. The following description is a brief synopsis. The algorithm has been described in more detail in [Horsch Poole, 1996] , and is similar to algorithms described in [Heckerman, Breese, Horvitz, 1989; Lehner Sadigh, 1993] For a given leaf l in a decision tree, its context fl l is extensible if it does not contain all the information variables. We refer to the information variables which are not in the context ....

[Article contains additional citation context not shown here]

Horsch, M. C., and Poole, D. 1996. Flexible policy construction by information refinement. In Proceedings of the Twelfth Conference on Uncertainty in Artificial Intelligence, 315--324.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC